BACKGROUND: Despite the proven value of mammography, its efficacy depends in large part on radiologists' interpretations and radiologists differ substantially in their interpretations. HYPOTHESES: Community mammography performance can be enhanced by better understanding sources of variability among radiologists and by working directly with radiologists in improving evaluation of their individual performance. SPECIFIC AIMS: 1.) To use statistical methods to estimate the accuracy of mammography at the level of individual radiologists and better understand reasons for variability. 2.) To better understand radiologist level characteristics associated with interpretive performance. 3.) To evaluate the feasibility and impact of an interactive web-based educational intervention and new audit reporting system. STUDY DESIGN: This cohort study will be organized within a conceptual framework previously proven to result in both physician behavior change and improved patient outcomes. Hierarchical statistical models will be developed using Breast Cancer Surveillance Consortium data from four population-based mammography registries in New Hampshire, Colorado, North Carolina, and Washington (Aim 1). Data from these Surveillance registries are available on more than two million mammography encounters. A survey, which will augment existing Surveillance data, will be sent to 400 radiologists to determine if radiologists understand concepts of numeracy (rates, risks and probability) in general and specifically related to breast cancer screening (Aim 2). Survey data will be linked with Surveillance data at an individual radiologist level. Information learned from Aims 1 and 2 will guide development and testing of an interactive web-based educational tool designed to improve radiologists' understanding of their own interpretive performance and to enhance use of audit reporting systems (Aim 3). RELEVANCE: Mandatory 'skills testing' is being considered in Congressional hearings as the Mammography Quality Standards Act is reviewed. Current methods of assessing the accuracy of individual radiologists are inadequate and there is no data to show that individual skills testing will improve accuracy of radiologists. This work builds on our previous model development and is a natural extension in a timely and important clinical area. [unreadable] [unreadable]